Legume abundance along successional and rainfall gradients in Neotropical forests

The nutrient demands of regrowing tropical forests are partly satisfied by nitrogen-fixing legume trees, but our understanding of the abundance of those species is biased towards wet tropical regions. Here we show how the abundance of Leguminosae is affected by both recovery from disturbance and large-scale rainfall gradients through a synthesis of forest inventory plots from a network of 42 Neotropical forest chronosequences. During the first three decades of natural forest regeneration, legume basal area is twice as high in dry compared with wet secondary forests. The tremendous ecological success of legumes in recently disturbed, water-limited forests is likely to be related to both their reduced leaflet size and ability to fix N2, which together enhance legume drought tolerance and water-use efficiency. Earth system models should incorporate these large-scale successional and climatic patterns of legume dominance to provide more accurate estimates of the maximum potential for natural nitrogen fixation across tropical forests. Data from 42 chronosequence sites show a geater abundance of legumes in seasonally dry forests than in wet forests, particularly during early secondary succession, probably owing to legumes’ nitrogen-fixing ability and reduced leaflet size.


Abstract
The nutrient demands of regrowing tropical forests are partly satisfied by nitrogen-fixing legume trees, but our understanding of the abundance of those species is biased towards wet tropical regions. Here we show how the abundance of Leguminosae is affected by both recovery from disturbance and large-scale rainfall gradients through a synthesis of forest inventory plots from a network of 42 Neotropical forest chronosequences. During the first three decades of natural forest regeneration, legume basal area is twice as high in dry compared with wet secondary forests. The tremendous ecological success of legumes in recently disturbed, waterlimited forests is likely to be related to both their reduced leaflet size and ability to fix N 2 , which together enhance legume drought tolerance and water-use efficiency. Earth system models should incorporate these largescale successional and climatic patterns of legume dominance to provide more accurate estimates of the maximum potential for natural nitrogen fixation across tropical forests.

Main
More than half of the tropical forest area worldwide is made up of secondary forests, which regrow after canopy removal due to natural or anthropogenic disturbances 1 . Second-growth forests are important globally because they supply firewood and timber, regulate the hydrological cycle, benefit biodiversity, and provide carbon storage as above-and belowground biomass 2,3 , but their growth can be constrained by nitrogen (N) availability 4 . Symbiotic fixation is thought to provide the largest natural input of N to tropical forests 5 , and part of the N demand of regrowing tropical forests is satisfied by legume trees (Leguminosae) that have the capacity to fix atmospheric N 2 through interactions with rhizobia bacteria 6 . The abundance of N-fixing legumes is not always strictly proportional to the rates of rhizobial activity, as some legumes downregulate fixation when the costs outweigh the benefits 7 . Nevertheless, legume abundance as represented by total basal area may provide a good estimate of the maximum potential N fixation in an ecosystem, with the advantage that this metric can be extracted from standard forest inventory surveys. Spatially explicit estimates of legume abundance through time could help to reduce uncertainties in Earth system models that include coupled carbon and N biogeochemistry 8 , but assessments of legume abundance have not yet been synthesized across the successional and climatic gradients that characterize tropical forests.
The abundance of N-fixing legumes relative to non-fixing trees has been closely examined in undisturbed tropical forests 9,10 and savannahs 11,12 . However, studies of legume abundance in regenerating forests are rare and have been restricted to the wet tropics 6,13,14,15,16 , so are likely not representative of tropical secondary succession globally. Due to environmental filtering 17 , systematic variations in legume abundance should occur along both rainfall and successional gradients. Wet and (seasonally) dry tropical forests 10,18 experience substantial differences in water and nutrient availability, which in turn may influence the competitive success of legumes in both biomes 19 . N-fixing legumes should have particular advantages in drier conditions; they can access N when mineralization rates decline due to low soil moisture 20 , and use their high foliar N content to maintain high growth rates and use water more efficiently relative to non-fixers 21 . Because young tropical forests are often N limited 4 , trees that are able to fix may be favoured during the earliest stages of forest regrowth after disturbance 22,23 . Some studies indeed show that legumes are more abundant in young compared with old wet Neotropical forests 6,14 , but others report successional trends in the opposite direction, with the relative abundance of these species instead increasing with stand age 13,15 . Characterizing these macroecological patterns of legume abundance across climate space and through successional time, along with variations in their functional traits, is crucial to determine whether our current knowledge of legume distributions can be generalized across the tropics and to achieve a more complete understanding of the role of this exceptionally diverse plant family within secondary Neotropical forest ecosystems.
Here we evaluate how the abundance of legumes (as measured by absolute and relative basal area) varies through secondary succession using data from 42 chronosequence sites 24 (Supplementary Fig. 1 and Supplementary Table 1) that span a large gradient in mean annual rainfall (from 750 to 4,000 mm yr −1 ) and explain legume success based on Nfixation potential and two functional traits related to drought tolerance (leaf size and leaf type). We primarily focused our climate analysis on mean annual rainfall ('rainfall' hereafter), but also tested the effect of rainfall seasonality 25 (the rainfall coefficient of variation from WorldClim) and climatic water deficit 26 (CWD; which tracks water losses during the months where evapotranspiration exceeds rainfall) as alternative predictors. Because leaf habit or leafing phenology is a better indicator of seasonal moisture stress than total annual precipitation 27 , we used this parameter to classify study sites as 'dry' forests if the vegetation was mostly drought deciduous (sensu ref. 10), or as 'wet' forests in all other instances (that is, mostly evergreen; Supplementary Table 1). As such, the terms 'dry' and 'wet' forests refer to two ecologically distinct tropical biomes with floristic compositions that differ in phylogenetic, biogeographic, functional and community ecological patterns (see refs 28,29,30 ). Therefore, although rainfall is used as the main (continuous) climatic variable to model legume abundance, we did not use this variable to classify sites as either dry or wet forests.
To understand the specific factors that enable legumes to thrive in particular tropical environments, we also investigated how the abundance of these trees related to their capacity to fix N 2 and a pair of crucial leaf traits that reflect drought tolerance. For each of the 398 legume species present at our sites, we assessed potential to fix N 2 based on positive nodulation reports and expert knowledge 31 (see Methods). Both of the leaf traits we examined-leaf size and leaf type-reflect adaptations to limited water availability (Supplementary Tables 2and 3). Smaller leaves have reduced boundary-layer resistance, which enables them to dissipate heat through conductive or convective radiation 32,33,34 . Leaf type is considered to be associated with drought severity and seasonality because plants with compound leaves (having either pinnate or bipinnate divisions) are able to shed individual leaflets (rather than whole leaves) when faced with severe moisture stress 35 . Our analysis demonstrates that the abundance of legumes indeed varies substantially and systematically across Neotropical forests, and although the ecological success (that is, high relative abundance) of these species during the very earliest stages of secondary succession is partly due to N fixation, other traits related to drought tolerance and water-use efficiency likely also offer competitive advantages.

Results
During the first three decades of forest regeneration, the total abundance of all legume trees as measured by their absolute basal area doubled in both dry and wet Neotropical forests (from 3 m 2 ha −1 in 2-to 10-year-old forests to 6 m 2 ha −1 in 21-to 30-year-old forests; Fig. 1a) as legume biomass gradually built up through succession. Here we define legume relative abundance (RA) as the basal area of Leguminosae trees divided by the total basal area of all trees in each plot and interpret it as a measure of ecological success that reflects legume performance relative to nonlegume species. Overall, although absolute legume abundance increased with succession, the RA of legumes declined with stand age in drier forests and declined with rainfall in younger forests (Table 1 and Fig. 1b). Despite these trends, site-to-site variation in successional change in legume RA was substantial ( Fig. 1b and Table 1). The fixed effects (stand age, rainfall and their interaction) accounted for 17% (marginal R 2 ) of the total variance explained by our model of legume RA, while 45% was due to site-to-site variation (conditional R 2 = 62%, Table 1). In the majority of dry forests, legume relative basal area decreased through time, which indicates legumes were initially a dominant component of early successional communities and then subsequently declined in abundance as other tree species became more common. By contrast, legume RA in the wet forest chronosequences typically began lower but remained constant through succession. The RA of legumes was much higher at the dry end of the rainfall gradient (rainfall effect, Table 1) and this difference was most evident during the first three decades of succession (0 to 30 years since abandonment, Fig. 1b). For example, in the youngest dry forests (2 to 10 years old), legumes on average made up more than one-third of the basal area of all trees (37%, compared with 18% in wet forests; Fig. 2a), and in some plots in Mexico (Chamela, Nizanda, Yucatán, Quintana Roo) and Brazil (Cajueiro, Mata Seca, Patos), relative abundance approached 100% (98% and greater). Although fewer chronosequences extend beyond three decades, in later successional stages (30 to 100 years old; Fig. 2d-f) legume abundance was still high in dry forests. The greater overall abundance of legumes in dry forests (compared with wet forests) may be partly a consequence of their higher initial recruitment, which is suggested by the high RA of small diameter legume trees during the first two decades of forest regeneration ( Supplementary Fig. 2). The RA of legume trees goes down during forest regeneration, but is markedly higher in seasonally dry forests than in wet forests, especially during the earliest stages of secondary succession. a, Plot-level total basal area of all legume species. b, Relative basal area of all legume species. c, Relative basal area of N-fixing legumes. d, Relative basal area of legumes with bipinnate leaves. Each circle represents one plot. Regression lines were drawn to highlight the successional trajectory of each of our 42 chronosequences. Dry forest sites (with dominant deciduous vegetation) are indicated in orange, and wet forests in blue. Insets show the average of all fits for absolute ( a) or relative (b-d) basal area of legumes in dry and wet forests. For forests in the wet Neotropics, the RA of legume trees is not influenced by variations in mean annual rainfall. However, below a threshold at approximately 1,500 mm yr −1 , legume RA increases as total rainfall diminishes. Results are stratified by stand age: a, 2-to 10-year-old forests; b, 11-to 20-year-old forests; c, 21-to 30year-old forests; d, 31-to 40-year-old forests; e, 41-to 60-year-old forests; f, 61-to 100-year-old forests. Results of a segmented linear fit are shown in each panel (** P < 0.01; ****P < 0.0001); N indicates the number of chronosequence sites with plots in each age category. Grey lines represent fitted values obtained using segmented regression models. Each circle represents site-level legume relative basal area averaged for each stand age category. Dry forest sites (with dominant deciduous vegetation) are indicated in orange, and wet forests in blue. Some dry forest sites receive higher average annual rainfall than some wet forest sites, but these sites are classified as 'dry' or 'wet' because of their functional composition (evergreen or deciduousness), not their mean climatology.
Mean annual rainfall was a strong determinant of legume RA over the entire Neotropical network (Table 1). Alternative models of legume RA that used rainfall seasonality and CWD as the main climatic predictor variable also explained a significant amount of variation in our data ( Supplementary  Figs. 3 and 4), but the best-supported model was based on mean annual rainfall (R 2 = 0.62, versus R 2 = 0.49 and R 2 = 0.48 for seasonality and CWD respectively; Supplementary Table 4). The magnitude of legume RA and its relationship with rainfall differed strongly between dry and wet secondary forests, most prominently during the first three decades of secondary succession (Fig. 2). For the 26 chronosequences from wet forests, mean legume RA was approximately 15% (±16% s.d.), within the range reported previously for individual sites 6,13,14,15,16 , and did not vary with rainfall. By contrast, legume RA in the 16 dry forest sites was much higher (41% ± 27% s.d.) and was strongly and inversely related to annual rainfall. The transition between these two patterns occurred at approximately 1,500 mm yr −1 (Fig. 2).
The functional traits of legumes also varied across the large-scale environmental gradients in our dataset. The spatial and successional patterns of legume abundance were largely driven by N-fixing species (Fig. 1c). For nearly two-fifths of the plots in our network, fixers were the only type of legumes present. At the plot level, the median percentage of total legume basal area comprised by fixers was 93.5%. The proportion of N-fixer basal area to total legume basal area did not vary with rainfall or stand age, and the RA of non-fixing legumes was much lower in both dry and wet secondary forest sites ( Supplementary Fig. 5). In contrast to the N fixers, the RA of non-fixing legumes remained constant throughout succession (Supplementary Table 5). When we stratified our analyses by leaf type, it was evident that the extremely high legume RA in young dry forests was largely due to the prevalence of species with bipinnate leaves (Figs. 1d and 3 and Table 1), which have significantly smaller leaflets than legumes with other leaf types (Supplementary Table 3). a-c, The exceptional ecological success of legumes in recently disturbed, waterlimited forests is mainly due to species that (1) are able to fix nitrogen and (2) have bipinnate leaves. Fitted values were obtained using a linear mixed-effects model, with stand age and rainfall as fixed effects and site as random effect (see (3)-(5) in Table 1 for full model results). Relative basal area and stand age were arcsin and logtransformed, respectively, before analysis. Models were computed separately for Nfixing legume species (a), legumes with bipinnate leaves (b) and legumes with pinnate leaves (c).

Discussion
Based on our survey of secondary forests across the Neotropics, we conclude the ecological success of legume trees is markedly higher in seasonally dry forests than in wet forests, especially during early stages of secondary succession. These findings agree with analyses of other large datasets from Africa and the Americas that found higher abundance of Nfixing legumes in arid conditions 12,36 , although those studies were unable to examine the effect of succession. We identified a threshold in mean annual rainfall at approximately 1,500 mm yr −1 -nearly identical to the threshold observed in forest inventories from North America 36 . Below this level, legume abundance was strongly and negatively correlated to water availability. Because this relationship was driven mainly by species that are both able to fix N 2 and have bipinnate leaves (Fig. 1), we suggest that the exceptional abundance of tropical legumes towards the drier end of the rainfall spectrum during secondary succession is the combined product of (1) small leaflet size, which allows for leaf temperature regulation and water conservation, and (2) N fixation, which contributes to photosynthetic acclimation, enhances water-use efficiency, and may satisfy the demand for nitrogen after the post-dry season leaf flush.
The availability of nitrogen is known to constrain biomass recovery in secondary Neotropical forests 4,37 because it is often lost following disturbance, either through harvesting of the canopy or crops, volatilization during burning or leaching 37 . Our finding that Neotropical legumes are proportionately more abundant in early succession throughout the Neotropics could be due to the high demand for N in recently disturbed forests 6,37 . Under those circumstances, fixation would allow legumes to overcome N limitation more easily than their competitors, although the degree to which initial secondary forest regrowth is limited by N availability is highly variable and influenced by local disturbance history and prior land use 38,39 .
Besides providing legumes with an advantage in young secondary forests in general, N fixation could offer additional benefits to plants growing under seasonally dry conditions. Forests that experience a pronounced annual dry season are affected by recurrent seasonal declines in soil moisture and, due to the associated hiatus in transpiration, plants are not able to access nutrients in the soil solution, including N (ref. 40). Because many dry forest trees lose their leaves each year and are required to grow an entirely new canopy, being able to obtain N through symbiosis could allow legumes to more quickly rebuild their leaf canopy at the end of the dry season compared with non-fixers 11,41 , which may need to wait until soil water is sufficiently recharged to acquire mineral N (ref. 40). Towards the high end of the Neotropical rainfall gradient, forests do not experience a strong seasonal moisture deficit, and as a result, are not usually faced with the regular nutrient scarcity that consistently affects dry forests. In addition to its role in satisfying seasonal nitrogen demands, fixation should also help legumes to further acclimate to hot, dry conditions 21,42 . By investing part of their fixed N into the production of photosynthetic enzymes, plants are able to draw down their internal concentrations of carbon dioxide, thereby creating steeper diffusion gradients in their leaves 43 . This adjustment allows photosynthesis to occur at lower stomatal conductance and reduced transpiration. These two advantages offered by fixation could account for the extremely high abundance of legumes early in dry forest succession, when air and soil temperatures are high, and soil water potential is at its lowest 42 , as well as their continued dominance in this ecosystem over evolutionary timescales 10,11 . But N fixation is clearly not the only trait that is advantageous to legumes in Neotropical dry forests. Because this biome is dominated by legumes that have bipinnate leaves particularly during the early stages of succession, it is clear that these species also benefit from their conservative use of water. Reduced leaf size (and the accompanying thinner boundary layer) offers a significant adaptive value in hot, dry environments 34 by enhancing heat dissipation and regulating leaf temperature more efficiently 32 . Many legumes also have the ability to adjust the angle of their laminae to regulate diurnal incident solar radiation 44 , which further reduces heat loading. Hence, having small leaflets could enable legumes at dry sites to benefit from high irradiance while preventing excessively high leaf temperatures. The bipinnate leaf type is confined exclusively to one subclade of legumes, the Caesalpinioideae 45 , and half of all caesalpinioid species in our dataset (mainly those that fix N 2 and were formerly classified as Mimosoideae 46 ) have bipinnate leaves. By contrast, only one-quarter of all N fixers have this trait, which means a relatively small subset of taxa is extremely well represented in the secondary forests of the dry Neotropics. This strong phylogenetic signal highlights the importance of drought tolerance traits as an adaptation to seasonal dry forests 11,47 , and demonstrates that leaf traits enhancing moisture conservation are equally as important as the potential to fix N 2 when explaining patterns of legume abundance. N-fixing legumes that have small bipinnate leaves thus hold a double advantage because those traits combine to minimize water loss during C acquisition and lead to increased water-use efficiency.
Across the Neotropics, a substantial amount of the variation in the relative abundance of legumes in secondary forests (17%) can be predicted by rainfall, stand age and their interaction. The strong negative relationship between legume RA and water availability was not sensitive to our choice of climate variable (either mean annual rainfall, rainfall seasonality or CWD; Fig. 2 and Supplementary Figs. 3 and 4), which suggests that legumes have a competitive advantage in dry climates that are characterized by recurrent seasonal droughts and strong water deficits during the growing season. The climate transition near 1,500 mm yr −1 , where the relationship between legume RA and rainfall switched from strongly inverse to flat, coincides with a known cross-over point in woody regeneration 48 . In regenerating dry forests, the canopy of established plants serves to moderate the harsh microclimatic conditions, thus facilitating the establishment of seedlings 48 , while in some wet forests, Nfixing legumes act to inhibit the growth of neighbouring trees during secondary succession 49 . A large fraction (45%) of the variance in legume abundance was associated with site identity (represented in our model as a random factor), which could be related to site-specific factors such as resource limitation (phosphorus, molybdenum, iron 50,51,52 ) or biotic interactions (dispersal, herbivory) that influence fixation in individual forest stands. More research is needed to determine how and to what extent these factors influence legume dominance at finer spatial scales.
N-fixing legume species attain their greatest levels of ecological success in Neotropical dry forests, where the benefits of fixation and co-occurring traits related to water conservation outweigh their costs throughout decades of succession. It is thus clear that insights about legumes derived from studies conducted exclusively in wet forests (for example, ref. 6) cannot be extrapolated across all Neotropical forests, and in particular are not valid for dry forests. As an alternative, future efforts to model the tropical N cycle must account for the strong heterogeneity exhibited by this hyperdiverse family of plants. Incorporating these large-scale abundance trends into Earth system models should allow for more accurate estimates of the potential for symbiotic N fixation across tropical forests. Our study also demonstrates that, even though the potential to fix N 2 through symbiosis is a crucial element of their success, it is not the only trait that accounts for the exceptionally high abundance of legumes in Neotropical forests. Leaf traits related to drought tolerance and water-use efficiency are also key adaptive features of dry forest legumes. In conclusion, our results provide a deeper mechanistic explanation for the abundance of Neotropical Leguminosae trees, which should be increasingly relevant to natural forest regeneration and ecosystem functioning as global temperatures warm and dry conditions become more widespread in the tropics 36,53 .

Chronosequence database
We extracted plot-scale legume abundance (m 2 ha −1 , basal area) from 42 previously published chronosequence studies 24 (2ndFOR network; Supplementary Fig. 1 and Supplementary Table 1). Our dataset includes lowland forests between 2 and 100 years old. Plot size varied from 0.008 to 1.3 ha, and across all 1,207 plots, mean plot area was 912 m 2 . The median number of plots per site was 14, ranging from 2 to 272. Prior land use in our sites included cattle ranching, shifting cultivation or a combination of the two. In each plot, all woody trees, shrubs and palms ≥ 5 cm in diameter were measured and identified, with the exception of sites in Costa Rica (Santa Rosa and Palo Verde) and Puerto Rico (Cayey) for which a minimum diameter at breast height of 10 cm was used. Across the network, mean annual rainfall varied from 750 to 4,000 mm yr −1 . Based on local site knowledge, study sites were classified as 'dry' forests when the vegetation is mostly drought deciduous, or 'wet' forests (mostly evergreen) in all other instances (sensu ref. 10; Supplementary Table 1). Because some tropical wet forests experience annual rainfall averages that overlap with the range exhibited by dry forests (particularly around 1,500 mm yr −1 ), we used seasonality in leaf habit (drought deciduous or evergreen) to distinguish between the two main biome types. We also repeated our analysis using two additional measures of water availability in the dry season, when water availability is a stronger limiting factor for plant growth and functioning: the rainfall coefficient of variation from WorldClim, which is an index of seasonality 25 , and the climatic water deficit (CWD, in mm yr −1 , defined as months where evapotranspiration exceeds rainfall 26 , http://chave.upstlse.fr/pantropical_allometry.htm), which reflects the relative intensity of water loss during dry months.

Functional traits
Across all sites, 398 legume species were present (Supplementary Table 2). We classified the potential of each species to fix N 2 based on positive nodulation reports for each species 31 (J.I.S. determined the fixation potential of the species not included in this reference). We determined leaf compoundness, which is considered an adaptation to severe moisture stress as plants are able to shed individual leaflets (rather than whole leaves). Because all legume species in our database had compound leaves, we also assessed the iteration of divisions, which we refer to as 'leaf type' (unifoliolate, pinnate or bipinnate). Leaf size reflects the thickness of the boundary layer and thus potential for heat dissipation, so we used representative collection specimens to measure length and width of legume leaflets (±0.001 cm). In total, we were able to quantify leaf size for 93% of all legume species within our dataset. To accurately represent the morphological variation of leaflets, for each leaf we averaged measurements made on leaflets from the bottom, middle and top of the axis. Leaflet size was measured on three leaves of each species (N = 3 individuals per species) using Neotropical specimens from the University of Minnesota Herbarium (133 of 398 species) or from online databases 54,55,56,57,58 . Leaflet length and width were highly correlated (R 2 = 0.82, P < 0.0001). Across our dataset, intraspecific variation in leaflet size was small compared with the proportion of variance explained by species (80% and 81% for leaf length and width, respectively), consistent with refs 34,59,60 . All N-fixation potential and leaf size data for each species are provided in Supplementary Table 3.

Legume abundance
We calculated (1) total legume basal area (m 2 ha −1 ) to serve as an approximate estimate of aboveground legume biomass, and (2) relative basal area (%), which summarizes the abundance of legumes compared with all species within each plot and is an ecologically meaningful indication of community composition. We interpreted legume relative basal area as a measure of legume ecological success because it reflects legume performance relative to other species. To determine whether N fixation and leaf type influence legume success, we also calculated relative basal area (%) for legume trees grouped by fixation potential and by leaf type.
Because the basal area of small diameter trees during early stages of forest regeneration is an approximation of recruitment, we stratified legume basal area by tree diameter classes. We focused our analysis on legume relative basal area as a measurement of legume relative abundance (RA) because it reflects biomass accumulation, but across our dataset, this variable was positively and significantly correlated to relative stem density (R 2 = 0.38, P < 0.0001).

Statistical analyses
To evaluate how legume abundance changed over successional time and along the rainfall gradient, we modelled legume RA as a function of stand age and mean annual rainfall with a linear mixed model using the lme4 package (v. 1.1.11) in R. We applied an arcsin and natural-logarithm transformation to improve the normality of RA and stand age, respectively. We included random site intercepts, as we expected between-site variation in initial legume abundance and random slopes to account for the variation of the effect of stand age on legume abundance among sites (Fig. 1). P values for the effect of stand age were calculated from the lmerTest package (v. 2.0.30), and random effect Pvalues were estimated via the likelihood ratio test. We obtained R 2 values for the effect of stand age (marginal R 2 ) and for the entire model (conditional R 2 ) using the r.squaredGLMM function in the MuMIn package 61 (v. 1.15.6). To determine whether other climatic variables also predicted legume abundance, we repeated models for legume RA using rainfall seasonality or CWD as the main climatic predictor variable. We compared the three models based on Akaike's information criterion adjusted for small sample sizes (AIC c ) and selected the best-supported model with the lowest AIC c . To determine the effect of rainfall on legume abundance at different stand age categories (2 to 10, 11 to 20, 21 to 30, 31 to 40, 41 to 60 and 60 to 100 years of forest age), we performed a piecewise linear regression using the Segmented package (v. 0.5.1.4). Lastly, to understand the successional change in legume basal area of trees of different size classes (<10 cm, 10 to 20 cm, 20 to 30 cm and >30 cm), we performed a multiple regression on arcsin transformed RA with stand age and forest type (dry or wet) as covariates. All analyses were performed in R version 3.2.2 (ref. 62 ).

Data availability
Plot-level legume basal area data from the 42 Neotropical forest sites are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.3p1k5d2.